\(\newcommand{\B}[1]{ {\bf #1} }\) \(\newcommand{\R}[1]{ {\rm #1} }\)

2022

View page source

Release Notes for 2022

mm-dd

12.19

1. In csv.join_file use the dictionary update method instead of the | operator because python 3.8 does not support the | operator.

2. csv.fit was failing when the USER environment variable was not defined. This has been fixed (none is the default user).

12-18

1. If a test fails, run it a second time with a different random seed. As the number of tests increases this helps avoid a by chance test failure.

2. Fix the value prior for iota in the sim_and_fit example. To be specific, it was log Gaussian by mistake and has been changed to Gaussian.

3. You should advance to dismod_at-20221218 to avoid a crash when all of the fixed effects have their lower and upper limits equal.

12-12

Improve the error messaging when both const_value and value_prior are empty in mulcov.csv .

12-08

1. Add the covariate_effect_depend_age_time option.

2. Add the option_sim.csv absolute_covariates option, and the option_fit.csv absolute_covariates option.

3. Also add the csv_absolute_covariate example. This can be used to help understand the covariate_effect_depend_age_sex option (as well as absolute covariates).

4. Remove the csv.fit wish list item. It was completed by the changes of integrand to integrand_name (see below).

12-07

1. The default value for db2csv and plot was true and should have been false. This has been fixed.

2. The csv simulate (csv fit) option file name was changed from option.csv (option_in.csv) to option_sim.csv ( option_fit.csv ) .

3. Change the csv fit option_out.csv file to option_fit_out.csv .

4. Add defaults for all the options in option_sim.csv .

5. Change the column node to node_name in fit_predict.csv and sam_predict.csv .

6. Change the column integrand to integrand_name in fit_predict.csv, sam_predict.csv and data_in.csv .

12-06

1. If the csv.fit parent_rate const_value was non-empty, the dage and dtime priors were treated as if they were empty. This has been fixed (even though a rate is fixed for one age and time, it might not be fixed for the next age and time).

2. The sim_and_fit test would randomly fail. Diagnostic information has been added (when the test fails) and the limits have been widened (to reduce the chance that the test fails).

3. The following csv.fit options were added: age_avg_split, ode_step_size.

12-03

1. Add the sim_and_fit example.

2. In the csv fit, extend the prior density to include the log_gaussian and log_laplace cases. This required adding the optional eta column to the prior.csv file. In addition, some of the other columns were made optional.

3. Add the csv fit quasi_fixed option.

12-02

1. Add the csv.join_file utility.

2. The meas_mean was not using the float_precision (because the corresponding value was a numpy.float64 instead of a float). This has been fixed.

3. Add a wish list item for csv.fit.

4. Fix bug in csv.fit where integrand names could be missing during prediction step.

11-28

Adapt to xrst-2022.11.28.

11-22

Include the no_ode fit in the csv fit optional db2csv and plot operations.

11-14

Add max_abs_effect to the csv fit option_in.csv file.

11-12

Add the refit_split option to the option all table and the option_fit.csv file. Note that its default value for the option all table is false, and for option_in.csv file its default value is true.

11-10

1. Add the random_depend_sex option to the csv simulation routine. In addition, add a discussion about option_sim.csv not being backward compatible.

2. Add the shared_memory_prefix option. In addition, add a discussion about option_fit.csv being backward compatible.

11-09

1. Add Boolean to the possible csv file data types.

2. Add the csv fit db2csv option. This required including a description of the dismod.db files in the csv fit documentation.

3. Add the csv fit plot option.

4. Add mention of the fact that the csv fit parent_rate.csv file corresponds to no random or fixed effects.

5. The following assert checked that the option file was not read twice:

       assert csv_option_value == None
   
   This has been removed because one python session may fit different cases,
   which could have different option values.
   

11-08

Fix and document using mulcov_id as an integrand_name in the predict.csv file.

11-03

Add number_sample to the option all table.

11-02

1. Check the csv.simulate Input Files and csv.fit Input Files to make sure they have all the necessary column names (if not print an informative error message).

2. The files fit_predict.csv and sam_predict.csv contained multiple copies of just one node instead of all the nodes that were fit. This has been fixed.

10-28

1. Remove the packaging wish list item because it was completed.

2. Add the Meta Regression wish list item.

10-27

1. Add the const_value column to the mulcov.csv input file.

2. Add a discussion of the different covariate multiplier types .

3. sex and one were added to the possible covariates in csv fit; see covariate .

10-25

1. Add the max_num_iter_fixed, tolerance_fixed, and max_number_cpu options.

2. Add an error message to csv fit when the root node sub-directory already exists in the fit_dir directory.

10-20

1. Remove the avgint Table item from the wish list (it was done previously).

2. Remove default None values for arguments to check_cascade_node and include argument type asserts in Syntax.

3. Change all_mtall to omega_all and remove all_mtspecific . In other words, do the calculation of mtall minus mtspecific (to get the omega constraint) outside of at_cascade. This completed and removed the mtall wish list entry.

4. In csv.fit, change all_predict.csv to sam_predict.csv. Make the column names in this file specialized for csv fit predictions. In addition, add fit_predict.csv to the csv fit predictions.

10-19

1. Add max_fit to the csv option table.

2. Change the definition of a Rectangular Grid so that it can be different for each value in the selection column.

3. Use the fact that the child rates (random effects) and the covariate multipliers are constant in age and time to simplify csv.fit. To be specific, use priors instead of smoothings for this functions (which enabled us to remove smoothing names from csv.fit); see parent_rate.csv, child_rate.csv, mulcov.csv.

4. There was an error in the all_predict.csv file. To be specific, the samples were not properly limited to the sex value corresponding to the fit. This has been fixed.

5. The covariate multipliers for sex are now frozen after the both fit at the root node level. The other covariate multipliers are frozen after the female and male split at the root level.

10-18

Change fit option.csv to option_fit.csv , add default value for all the fit options, and place resulting options in option_fit_out.csv .

10-17

First tested version of csv.fit .

10-16

Add the csv.empty_str utility for converting empty strings to, and from, None.

10-15

Change license notice to SPDX format.

10-14

1. Improve the description of the location of the fit results; see Output dismod.db .

2. Add the mtall_data and avgint Table wish list entries.

3. Fix bug in csv.simulate . To be specific, one of the rate functions was being used for all the rates. A regression test for this bug was added (where is more than one rate in no_effect_rate.csv ).

10-10

Fix the return value for spline_dict function calls:

z = spline_dict[z_name](x, y)

so that the result z is a float and not a numpy array of size one.

10-09

1. Move the csv interface files to the csv sub-module; see csv.module.

2. Move

   • read_csv_table to csv.read_table

   • write_csv_table to csv.write_table

10-06

Fix some documentation spelling errors detected by new version of pyspellchecker.

09-18

1. Adapt to xrst-2022.9.18. This changes the location of documentation html files from the doc/rst directory to the doc directory.

2. Change the linking text for web pages from the page title to the page name. For example, use all_node_db instead of The All Node DataBase to link to the all node database documentation.

08-27

1. Remove the relative_tolerance option from csv.simulate and increase absolute_tolerance in its example. (It was removed from dismod_at.average_integrand because testing it with csv.simulate_xam revealed problems.)

2. Remove \begin{align*} from math directives (sphinx automatically adds a split to math directives).

08-25

Add the absolute_tolerance, relative_tolerance, and float_precision columns to the option_sim.csv file.

08-23

1. Add tracing to csv.simulate ; i.e., it will report is progress.

2. Change number of spaces corresponding an indent from 4 to 3. (This interacts better with the sphinx documentation system.)

08-21

1. Fix some error message and add some tracing to csv.simulate .

2. Update documentation to use xrst-2022.8.21.

08-16

Add the section ode_iota_omega which derives the dismod ODE solution when rates are constant and only iota, omega are non-zero.

08-15

Remove use of the python distutils package because it has been deprecated.

08-12

Add the bilinear spline routine.

08-09

1. Advance to xrst-2022.8.8 (this improves the documentation).

2. Improve the index using the keyword file to exclude some entries .

3. Add the csv.read_table and csv.write_table routines to the at_cascade module.

08-01

Start construction of a cascade data simulation tool; see csv.simulate .

04-18

Remove shared memory and LD_LIBRARY_PATH items from the wish_list because they have been completed.

03-22

Add balance_fit to the option_all table.

03-14

1. Fix a bug in how avgint_parent_grid and create_shift_db pass priors down the cascade. To be more specific, the split reference covariate effects were not being handled properly.

2. A check of the prior mean for iota was added at the end of the split_covariate example

3. Add the fit_type argument to no_ode_fit.

03-12

The shared memory names were shortened, to be specific _at_cascade was removed from theses names. In addition, the all node database was added as an argument to clear_shared so it would know the shared memory prefix.

03-10

Add the fit_type_list argument to cascade_root_node and continue_cascade.

03-08

The remission example was added.

03-07

The shared_memory_prefix was added to the option_all table.

03-05

Change no_ode_fit to always use Gaussian data densities.

02-05

Use com_cov_reference to compute references for relative covariates and remove the all_cov_reference table from the all_node_db.

02-04

In the examples, fix the following typos: Change ‘are checked for check_cascade_node’ to ‘are checked by check_cascade_node’. Change ‘rood_node_db routine’ to ‘root_node_db routine’.

02-03

Add new wish_list entries and remove old ones the have been completed.

01-28

Add the add_log_entry, move_table, and empty_avgint_table routines to the at_cascade module.

01-27

1. The max_fit value is now doubled when doing mulcov_freeze fit.

2. If max_number_cpu was one, the optimizer trace was not printed to the screen. This has been fixed.

01-26

Change perturb_optimization_scaling to perturb_optimization_scale and add the perturb_optimization_start option.

01-25

The clear_shared routine was added to the at_cascade module.

01-08

1. Fix a bug in create_shift_db. To be specific, the mean int a value prior child variable in the shifted database (child database) might not satisfy the prior limits. This violated the dismod_at assumptions about the value prior means.

2. Change dage and dtime Priors to be difference predicted by parent fit.

01-02

1. The cascade would hang at the end, instead of detecting completion, when max_number_cpu was greater the one. This has been fixed.

2. Use job_name as the key for each job in the message_dict return value.